By Nils Ekström, CTO at Stomme AI
Every chatbot you've used works the same way: you open a browser tab, type a prompt, get a response, close the tab. Your conversation history, your preferences, your context — all stored on someone else's servers. The AI company decides what to keep, how long to keep it, and what to do with it.
For quick questions, that's fine. But when an AI handles your email, your calendar, your client work, your business operations — you need a different architecture.
What "local-first" actually means
Let's be precise about what runs where, because most "local AI" claims are misleading. Here's exactly how a stomme.ai agent works:
On your Mac (local): The agent itself — its workspace, memory, conversation history, files, connected tools, orchestration, scheduling, and all accumulated context. This is the infrastructure that makes an agent an agent rather than a chatbot. It runs as a background service on your Mac, like any other application.
In the cloud (API calls): AI reasoning. When your agent thinks — drafts an email, analyses a document, plans a project — it sends a request to Anthropic's Claude API and receives a response. This is the same API that powers Claude.ai. Anthropic's terms prohibit using customer data for model training.
On our servers: Billing, onboarding, and account management only. We don't see your conversations, your files, or your agent's work.
The honest version: your agent's brain uses cloud AI. Your agent's body — everything it knows, remembers, and does — lives on your Mac.
Why this matters more than "fully local"
Some people hear "cloud AI" and think it defeats the purpose. But consider what an agent accumulates over months of work: your communication patterns, your project context, your client relationships, your preferences, your working style. That's the valuable data. That's what stays on your machine.
The AI reasoning is stateless — each API call is a question and answer. Nothing is stored on Anthropic's side between requests. Your agent sends context, gets a response, and the context goes back to your local workspace.
This is fundamentally different from a cloud chatbot, where everything — history, context, preferences — lives on their servers.
What you actually own
When you stop subscribing to a chatbot, your conversation history disappears (or becomes inaccessible). With a stomme.ai agent, everything your agent has built — its memory, its project files, its accumulated understanding of your work — stays on your Mac as ordinary files. You can read them, back them up, or migrate them.
After six months of handling your operations, that accumulated context is genuinely valuable. It's a file on your hard drive, not hostage to a subscription.
Privacy you can verify
Your agent's workspace is a folder on your Mac. You can open it and read every file. You can monitor network traffic and see exactly what goes where. You don't need to trust a privacy policy — you can inspect the architecture.
For businesses handling client data, regulated information, or competitive intelligence, this matters. Your agent's knowledge base — the sensitive part — never leaves your network. Only individual reasoning requests go to the AI provider, and those are covered by Anthropic's commercial API terms (no training on your data).
Can you run fully local models?
Yes. If you have capable hardware (Apple Silicon Macs work well), you can configure your agent to use local AI models instead of or alongside cloud APIs. This gives you fully offline reasoning with no data leaving your machine at all — at the trade-off of smaller models. It's a configuration choice, not an architectural limitation.
The bottom line
We don't claim your data "never leaves your device" — that's a marketing line most companies can't back up honestly. We tell you exactly what runs where: your agent infrastructure is local, AI reasoning uses cloud APIs, and your accumulated data stays on your Mac.
That's a meaningful difference from every chatbot, even if it's not the simple "100% local" story some people want to hear.
Your agent runs on your Mac. AI reasoning via cloud API. Your data stays yours.